BAD NEWS, QUALIFIED

Hmmm, here's a poll that looks bad, from AP-GfK: McCain and Palin are beating Obama and Biden, with a lead of 4 points (48%-44%) among likely voters.

But here's the stat (from a PDF of the poll results) that makes me suspicious of the results (emphasis added):

Earlier you mentioned that you had voted in the 2004 elections. Did you vote for George W. Bush, the Republican, John Kerry, the Democrat, or someone else?
Likely Voters
George W. Bush 50
John Kerry 40
Other 7
Don’t know 1
Refused 3

Er, the 2004 popular vote percentages were Bush 51%, Kerry 48%, Nader 1%.

So how is this a representative sample?

This isn't the first recent poll to show a surprising skew in party composition, as Seth Colter Walls recently noted at the Huffington Post. FiveThirtyEight.com's Nate Silver says this isn't the result of sinister machinations by pollsters -- it's possible there actually is some sort of party shift taking place, but it may just be that, since the Republican convention, people who've voted Republican in the past are now more willing to talk to pollsters.

In any case, a poll with a pro-GOP skew that large -- and a much smaller lead for McCain -- suggests to me that Obama is hanging in there with regard to the popular vote.